Global DNA Hypomethylation in Peripheral Blood Mononuclear Cells as a Biomarker of Cancer Risk

DNA methylation is an epigenetic phenomenon that affects the regulation of gene expression and genome integrity (1). Aberrant DNA methylation is a critical mechanism for carcinogenesis (2, 3). Global DNA hypomethylation is an almost universal finding in many human cancer tissues (4) and cancer precursor cells (5), and many studies in rodent models showed that global DNA hypomethylation can, by itself, induce cancer (6). However, it is yet unclear whether this common epigenetic feature can be a clinically functional biomarker or even a screening tool to identify patients affected by cancer disease. A stimulating scientific debate is indeed ongoing to clarify the usefulness of genomic methylation status in DNA obtained from an easily accessible tissue in humans such as peripheral blood mononuclear cells (PBMC) as a suitable biomarker even for cancer tissue of different origin (7).

Differently from mutations and LOH, DNA methylation is a reversible phenomenon (8, 9) and it can be modified by nutrients such as folate (10) as well as influenced by the common 677C>T polymorphism in methylenetetrahydrofolate reductase (MTHFR) gene (11, 12). An interaction between folate status and the MTHFR variant can affect global DNA methylation (11), and thereby potentially modify the risk of cancer. The polymorphic MTHFR677C>T favors the reduction of the 5-methyltetrahydrofolate (5-methylTHF) proportion in tissue folate pool (11), and at the low folate status individuals carrying the MTHFR677C>T homozygous variant genotype (677TT) have a decreased global DNA methylation (11, 12), which may induce molecular modifications in the cell eventually leading toward the development of cancer (13). Most recently, DNA methylation has been tested in blood as a circulating tumor cell DNA marker (14) and a number of studies evaluated the possible role of circulating white blood cells DNA methylation in different types of cancer as a potential marker to define the risk for malignancies of different tissue origin (7, 15–18).

We, therefore, set out to determine whether global hypomethylation in PBMCs DNA can indicate the presence of cancer or impending cancer development.

From an original sample-set of 753 subjects of both sexes and mean age of 64.8 ± 7.3 years, who were recruited from a single geographic area in Italy with a similar socioeconomic background in which they were previously engaged from 1996 to 2004 for a cardiovascular risk factors study (19, 20), we retrospectively evaluated the clinical history for cancer diagnosis and selected 68 subjects who were diagnosed of cancer at the time of enrollment in the study and 62 subjects who developed cancer during an 8-year follow-up period, along with 68 age- and sex-matched controls at enrollment and 58 age- and sex-matched controls, at follow-up. Clinical history of the subjects referring to cancer disease was obtained by search within the national population register, and by an ambulatory or a telephone-administered survey conducted by a physician. The causes of death including cancer were obtained from death certificates kept at the Italian Institute of Statistics (ISTAT; Rome, Italy). This study was approved by the University Hospital of Verona Ethical Review Boards and informed consent was obtained from all subjects after full explanation of the study.

Samples of peripheral venous blood were drawn in EDTA-containing BD Vacutainer tubes from each subject after an overnight fast, before proceeding for plasma as well as DNA extraction. Plasma was accurately stored at −80°C before analyses.

DNA was extracted from PBMCs by Wizard Genomic DNA Purification Kit (Promega Corporation) and global DNA methylation, expressed as percentage (%) 5-methylcytosine (mCyt)/(mCyt + Cyt), was determined using a liquid chromatography/mass spectrometry (LC/MS) method, as we previously described (11, 21). Percentage relative SDs for method precision of 1.9 (within-day) and 1.7 (day-to-day; ref. 21). Plasma folate concentrations were measured by an automated chemiluminescence method (Chiron Diagnostics) having an intra-assay and interassay coefficients of variation of 5% and 9%, respectively. Total plasma homocysteine (tHcy) was measured by a high-performance liquid chromatography (HPLC) method with fluorescent detection, as previously described (11). High-sensitivity C-reactive protein (hs-CRP) was measured by a particle-enhanced nephelometric immunoassay with commercially available methods in a BNII Behring Nephelometer Analyzer (Dade Behring Inc.), as previously described (22).

The analysis of the MTHFR677C>T polymorphism was conducted by PCR followed by HinfI digestion (19, 23).

The distribution of continuous variables in groups are expressed as mean ± SD. Statistical analysis was conducted with log-transformed data for all skewed variables and geometric means (antilogarithms of the transformed means) are presented with 95% confidence intervals (CI). Statistical significance for differences in continuous variables was tested by Student unpaired t test or ANOVA variant analysis, with polynomial contrasts for linear trend when appropriate. Categorical variables were analyzed using a χ² test, with χ² for linear trend when appropriate. Adjustment for confounding variables was conducted by general linear model or logistic regression analysis. Interaction terms were estimated by means of general linear models. Receiver operating characteristic (ROC) curve analysis was conducted to identify the most efficient cutoff value of DNA methylation for cancer diagnosis as well as incident cancer occurrence during follow-up. The plasma folate concentrations corresponding to the 50° percentile in controls (12 nmol/L) were used as the threshold for either low (<12 nmol/L) or high folate (≥12 nmol/L). Statistical significance refers to a two-tailed analysis in which P value is less than 0.05.

The prevalence of cancer in this cohort was 9.2%, which is similar to that reported for the whole population in the same Italian geographic area. Cancer types were hematologic malignancies (20%), bladder cancer (19%), gastrointestinal cancers (13%), prostate (12%), breast (9%), kidney (7%), lung (6%), and larynx cancers (4%).

Cancer cases and controls as compared with other subjects within the study cohort showed a higher prevalence of coronary artery disease-affected subjects (88.4% vs. 69.8%; P < 0.001), slightly higher age (64.52 ± 10.89 vs. 59.74 ± 7.27; P < 0.001), prevalence of smokers (60% vs. 73%; P = 0.01), and subjects of male sex (83.7% vs. 75%; P = 0.039). No differences were instead detected in terms of plasma tHcy, folate, hs-CRP, and vitamin B12 concentrations (P > 0.2 for all the comparisons).

The main clinical and biochemical characteristics of patients with or without cancer history at baseline are described in Table 1. As shown, there were no significant differences between cancer cases and controls as it referred to age, sex, smoking habit, inflammatory marker hs-CRP, tHcy, and vitamin B12. Plasma folate concentrations were significantly lower in cancer subjects than in controls (9.86 nmol/L; 95% CI, 8.26–11.77 vs. 13.24 nmol/L; 95% CI, 11.9–14.72; P = 0.003). The frequency of homozygous mutants for the MTHFR677C>T polymorphism was significantly higher in cancer subjects as compared with controls (31.03% vs. 11.70%; P = 0.013). Moreover, cancer subjects who already had cancer at enrollment showed reduced global DNA methylation compared with controls who did not have cancer at enrollment (4.39%; 95% CI, 4.25–4.53 vs. 5.13%; 95% CI, 5.03–5.21, expressed as percentage mCyt/mCyt+Cyt; P < 0.0001; Table 1). By stratifying the study population according to quartiles of DNA methylation (defined on the basis of control's levels), subjects with cancer history were clearly more represented within the lowest quartile as compared with the others (P < 0.001 by χ² test; Fig. 1). To determine a precise cutoff to differentiate cases from controls, a ROC curve analysis was conducted using data from the overall population, on the basis of which a threshold level of 4.74% for DNA methylation was chosen [area under curve (AUC) = 0.880 with 95% CI, 0.817–0.944; Fig. 2]. Subjects with DNA methylation lower than 4.74% showed a much increased prevalence of cancer (43 of 47) than those with higher (≥4.74%) DNA methylation levels (91.5% vs. 19%; P < 0.001 by χ² test; OR, 45.9 with 95% CI, 14.3–147.6; Fig. 3). Such association persisted statistically significant even after adjustments for sex, age, smoking habit, serum concentrations of inflammatory marker hs-CRP as well as plasma folate, vitamin B12, and tHcy concentrations (OR, 35.9 with 95% CI, 8.0–160.5; Fig. 3).

In a gene–nutrient interaction model analysis, to better define the potential significance of the DNA methylation cutoff value, the combined effect of MTHFR677C>T genotype and plasma folate concentrations in determining DNA methylation levels and in relation with cancer history was explored (Fig. 4). No statistically significant interaction term was found (for DNA methylation P = 0.718; for cancer history P = 0.201). However, the data were indicative for an additive effect between MTHFR677C>T genotype and plasma folate concentrations in determining DNA methylation, with the subgroup of MTHFR677TT homozygous carrier associated with low plasma folate concentrations having the lowest DNA methylation levels (4.39% mCyt/mCt+Cyt for MTHFR677TT with low folate; Fig. 4A). The reduction in DNA methylation levels associated with MTHFR677T allele was significant only in those subjects with low plasma folate concentrations (P = 0.019 by ANOVA with polynomial contrast for linear trend), whereas among those with high plasma folate concentrations there was no statistically significant association (P = 0.181 by ANOVA with polynomial contrast for linear trend). Consistently, the subgroup of MTHFR677TT homozygous carrier associated with low plasma folate concentrations showed also the highest prevalence of subjects with cancer history (81.2% MTHFR677TT vs. 52.2% MTHFR677CT vs. 38.1% MTHFR677CC; P = 0.011 by χ² for linear trend). The association between MTHFR677C>T genotype and cancer was significant only in subjects with low plasma folate concentrations (P = 0.011 by χ² for linear trend) and not among those with high plasma folate concentrations (P = 0.761 by χ² for linear trend). Considering MTHFR677CC subjects with high folate concentrations as reference group (OR = 1), only MTHFR677TT subjects with low folate concentrations showed a higher risk of cancer (OR, 7.04; 95% CI, 1.52–32.63; P = 0.013), whereas no significant association was observed for MTHFR677TT subjects with high folate (OR, 1.77, 95% CI, 0.53–5.89; P = 0.35; Fig. 4B).

The main clinical and biochemical characteristics of patients with cancer during follow-up as compared with controls are described in Table 2. Controls and incident patients with cancer were age- and sex-matched and with similar percentage of subjects with smoking habit. As shown, DNA methylation at time of enrollment was lower in incident patients with cancer as compared with controls [4.47% (95% CI, 3.92–5.12) vs. 5.07% (95% CI, 4.95–5.18); P = 0.001]. Among those analyzed for DNA methylation, 10 subjects died during the follow-up period: 4 of them died of cancer and 6 of other causes, mainly of cardiovascular disease. Those who died of cancer had a lower global DNA methylation as compared with those who died of other causes [4.17% (95% CI, 3.48–4.96) vs. 4.89 (95% CI, 4.57–5.19); P = 0.015]. The survivors had overall a significantly higher methylation as compared with those who died of cancer [4.93% (95% CI, 4.79–5.01) vs. 4.17 (95% CI, 3.48–4.96); P = 0.004] but not to those who died of other causes [4.93% (95% CI, 4.79–5.01) vs. 4.89% (95% CI, 4.57–5.19); P = 0.79]. A further analysis was carried out, to determine a precise cutoff to differentiate subjects who did or did not develop cancer during follow-up, a ROC curve analysis was conducted, on the basis of which a threshold level of 4.69% for DNA methylation was chosen (AUC = 0.834 with 95% CI, 0.714–0.954). Subjects with DNA methylation lower than 4.69% showed a much increased occurrence of incident cancer than those with higher (≥4.69%) DNA methylation levels (66.7% vs. 5.3%; P < 0.001 by χ² test; OR, 36.0 with 95% CI, 3.9–329.9). Such association persisted statistically significant even after adjustments for sex, age, smoking habit, serum concentrations of inflammatory marker hs-CRP as well as plasma folate, vitamin B12, and tHcy concentrations (OR, 43.6 with 95% CI, 2.7–708.4).

DNA methylation has been extensively studied as an epigenetic mechanism for carcinogenesis as well as a potential biomarker for cancer disease. However, decades of study could not clarify the mechanistic and clinical significance of DNA methylation due to its diverse biologic functions as well as its unique tissue specificity. In the present study, we enlightened the value of DNA methylation as a clinical index to screen patients affected by cancer or at risk to develop cancer disease. Interestingly, global DNA methylation status of cancer subjects was invariably decreased compared with control group subjects, indicating that cancer subjects have a systemically decreased global DNA methylation. Furthermore, we also observed that global DNA methylation in PBMCs at the enrollment was lower in subject who developed cancer during the 8-year follow-up period as compared with those who did not develop cancer during the same period.

Our observations are consistent with previous studies, which investigated the relationship between peripheral blood leukocytes DNA methylation and the risk of cancer (24). In a case–control study, leukocyte DNA hypomethylation was associated with increased risk for adenoma (P_trend = 0.01) and a nonsignificantly increased risk for cancer (P_trend = 0.08). In that study, colorectal cancer subjects also had significantly lower blood folate (16). In the Colorectal Neoplasia Screening With Colonoscopy in Asymptomatic Women at Regional Navy/Army Medical Centers (CONCeRN) Study subjects, women in the second (OR, 0.72; 95% CI, 0.34–1.52) and third tertiles (OR, 0.17; 95% CI, 0.06–0.49) had lower risk of colorectal adenoma, which is a precursor of colorectal cancer, compared with women in the lowest tertile of leukocyte genomic DNA methylation (P_trend = 0.002). Several other studies in which the overall white blood cell genomic DNA methylation was measured as a marker of different cancer types including bladder (25–27), stomach (17), breast (28), and head and neck cancer (18), have found an elevated risk for cancer among those in the lower quantile of genomic DNA methylation compared with those in the highest quantile, regardless of the methods to measure global DNA methylation.

Owing to the LC/MS DNA methylation assay that can precisely measure DNA methylation status, it was possible to observe cutoff values, interestingly very similar in both prevalent and incident cancer study design (4.74% and 4.69% global DNA methylation, respectively), that clearly differentiate controls from cancer subjects by DNA methylation status, indicating that this values might be considered as a clinically valuable marker for cancer screening. Our study also indicates that a possible cause of conflict results from previous studies that determined the efficacy of global DNA methylation in PBMCs as a cancer biomarker might be the method used to measure DNA methylation, which does not always have adequate sensitivity and reproducibility.

Even though we found a significantly lowered genomic DNA methylation status in patients with cancer, it is not straightforward to explain why patients who have cancer originated from different types of organs have a decreased DNA methylation status in their PBMCs, especially because epigenetic patterns are usually considered to be highly tissue specific. Furthermore, PBMCs are normal cells that are expected to maintain normal DNA methylation status as compared with cancer cells, which instead tend to have decreased global DNA methylation. We, however, may speculate about the implication of several mechanisms by which cancer subjects have decreased global DNA methylation in PBMCs: (i) cancer developed from a certain organ may alter the whole body metabolism such as energy metabolism or one-carbon metabolism that mediates methyl transfer reactions even for maintenance of a stable DNA methylation status. The global metabolic unbalance as well as that of DNA methylation reaction in cancer may subsequently induce a reduction in global DNA methylation of PBMCs; (ii) circulating cancer cells or cancer products such as miRNAs or inflammatory mediators directly affect the PBMCs DNA methylation in blood; and (iii) the systemic condition that accelerates cancer development concurrently reduces DNA methylation in PBMCs.

The results of this study support the hypothesis that decreased global DNA methylation status rather indicates a systemic condition prone to cancer development instead of a direct cancer effect, because: (i) in our study, global DNA methylation in PBMCs was already decreased before developing cancer, and (ii) findings from this study show a combined effect between MTHFR677C>T genotype and plasma folate concentrations, both of which have been regarded as cancer risk factors, for DNA methylation levels and cancer risk at the same time: if associated with low plasma folate, the MTHFR677TT homozygous mutant genotype seems to be crucial in determining the lowest DNA methylation levels and the highest risk of cancer.

Because DNA methylation is a predictor for the development of cancer, it seems that lowered DNA methylation in PBMCs imply that this epigenetic feature collectively reflects the systemic condition that can favor the development of cancer. In this regard, it would be of high interest to follow-up the patients for a longer period and to enlarge the population sample set to be able to evaluate the drop in global methylation levels according to different types of cancer as well as to timing of cancer development. Further studies are warranted in this view.

Among the most interesting findings of the present study, there was also the observation of a decreased DNA methylation status related with lower folate status and higher frequency of MTHFR677TT genotype. Furthermore, MTHFR677TT variants further increased the risk only in those subjects with low folate status (OR for cancer of 7.042; 95% CI, 1.52–32.63; P = 0.013). These observations are consistent with previous reports describing that higher cancer risk associated to the presence of MTHFR677TT genotype only for those subjects with low folate status (29) but the link with reduced methyl availability due to lower plasma folate concentrations was previously only hypothesized (29), and this was confirmed by the present data in which a potential mechanism for this relationship is brought into play by the correlation with lower DNA methylation status.

In conclusion, the low global DNA methylation levels in PBMCs may be a predictor of cancer. The low global DNA methylation can also predict the development of cancer in the near future. Measurements of plasma folate concentrations and genotyping for MTHFR677C>T polymorphism may give an additional impact in subjects with low DNA methylation status. In the future, larger prospective clinical studies are certainly warranted. Whether differences in DNA methylation are related to disease status or outcomes as well as the understanding of precise correlation between PBMCs and specific cancer tissue DNA methylation also need to be determined.

No potential conflicts of interest were disclosed. Any opinions, findings, conclusion, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the view of the U.S. Department of Agriculture.

Conception and design: S. Friso, S.-W. Choi

Development of methodology: S. Friso, P. Guarini, P. Pattini, S.-W. Choi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Friso, S. Udali, P. Guarini, C. Pellegrini, P. Pattini, S. Moruzzi, D. Girelli, F. Pizzolo, N. Martinelli, O. Olivieri, S.-W. Choi

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Friso, S. Udali, S. Moruzzi, F. Pizzolo, R. Corrocher, O. Olivieri, S.-W. Choi

Writing, review, and/or revision of the manuscript: S. Friso, S. Udali, P. Guarini, C. Pellegrini, S. Moruzzi, D. Girelli, F. Pizzolo, R. Corrocher, O. Olivieri, S.-W. Choi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Friso, C. Pellegrini, P. Pattini, O. Olivieri

Study supervision: S. Friso, N. Martinelli, R. Corrocher, S.-W. Choi

The authors thank Mrs. Maria Zoppi for her invaluable help for the secretariat work and Mr. Diego Minguzzi for his excellent technical assistance.

This material is based on work supported by the U.S. Department of Agriculture, under agreement No. 51000–074-01S. This project has been supported in part by the NIH Grants R01 AG025834 (to S.-W. Choi) and in part by Cariverona Foundation, Italy (to S. Friso).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.